Rheology Modelling of Cement Paste with Manufactured Sand and Silica Fume: Comparing Suspension Models with Artificial Neural Network Predictions

Manufactured sand is increasingly used in concrete and predicting the rheology of such suspensions based on their composition are necessary. In this study, emphasis is on cement pastes with manufactured sand and silica fume. An artificial neural network, namely multilayer perceptron, is compared with nine suspension models: two liquid thickness models, the relative concentration of solids and six relative viscosity models based on the relative concentration of solids. Measurements on 107 mixes with filler (0–125 µm) from manufactured sand are conducted, acquiring yield stress, plastic viscosity, flow resistance ratio and mini slump flow. None of the suspension models offer good correlations to the measured parameters for all mixes, but an increase in prediction accuracy is seen for subsets of materials, especially mixes without silica fume. The artificial neural network outperforms the suspension models, providing a coefficient of determination between 0.84 and 0.91 for all mixes, thus illuminating a new pathway for cement paste rheology modelling.